07 Erstellung von Grafiken mit ggplot2

Überblick

Das Paket ggplot2 ist die meistgenutze Grafikbibliothek in R. Sein modularer Aufbau in aesthetics, coordinates und geometries erlaubt beliebige Freiheit in der Gestaltung von Plots.

Dataset

Datensatz midwest aus dem Package ggplot2 enthält Daten einer Volkszählung.

data(midwest)
# Umwandlung einiger Variablen in Datentyp 'factor'
midwest <- midwest %>% mutate_at(vars(county, state, inmetro, category), as.factor)
# Show sample
sample_n(midwest, 10) %>% DT::datatable(width = 700, options=list(scrollX = TRUE))

Histogram

Verteilung einer numerischen Variable

midwest %>% 
    ggplot(aes(x = percasian / 100)) + 
    geom_histogram(bins = 50, color=1, fill='#d77e2d') +
    scale_x_continuous(labels=scales::percent) +
    labs(x='Percent asian of tot. pop.', y = 'Count', 
        title='Distribution of Asian percentage per district')

Quantilsplot

Vergleich einer Verteilung mit Normalverteilung

midwest %>% 
    ggplot(aes(sample=percasian)) + geom_qq() + stat_qq_line()

midwest %>% 
    ggplot(aes(sample=log10(percasian))) + geom_qq() + stat_qq_line()

Density plot

Verteilung einer numerischen Variable im Vergleich mit wenigen Kategorien

midwest %>% 
    ggplot(aes(x = percasian / 100, fill = inmetro)) + 
    geom_density(alpha=.5) +
    scale_x_continuous(labels = scales::percent) +
    scale_fill_discrete(labels = c('Yes', 'No')) +
    labs(x='Percent asian of tot. pop.', y = 'Count', 
        title='Distribution of Asian percentage per district',
        fill = 'In metro area')

Boxplot / Dotplot

Verteilung einer numerischen Variable über mehrere Kategorien

midwest %>% 
    ggplot(aes(x = state %>% fct_reorder(percollege), y = percollege, fill=state)) + 
    geom_dotplot(binaxis = 'y', stackdir = 'center', dotsize=.6) +
    geom_boxplot(alpha = .3, outlier.size = 0) +
    labs(x = 'State', y = 'College education in percent of tot. pop.')

Scatterplot

Relation zwischen zwei numerischen Variablen

midwest %>% 
    ggplot(aes(x=area, y=poptotal)) + 
    geom_point(alpha=.5, color='#a52dd7') + 
    geom_smooth(method="lm") +
    scale_y_log10(labels= scales::comma) +
    labs(x = 'Area', y = 'Total population', 
         title='Population per Area', 
         subtitle = 'in log scale')

Als facet-plot unterschieden nach state

midwest %>% 
    ggplot(aes(x=area, y=poptotal)) + 
    geom_point(alpha=.5, color='#a52dd7') + 
    geom_smooth(method="lm") +
    scale_y_log10(labels= scales::comma) +
    labs(x = 'Area', y = 'Total population', 
         title='Population per Area', 
         subtitle = 'in log scale') +
    facet_wrap(vars(state))

Matrix-Scatterplot

library(GGally)

midwest %>% 
    select(percollege, percbelowpoverty, percblack, percasian, inmetro) %>% 
    GGally::ggpairs(mapping=aes(color=inmetro))

Barplots

midwest %>% 
    ggplot(aes(state, fill = inmetro)) + geom_bar()

midwest %>% 
    ggplot(aes(state, fill = inmetro)) + geom_bar(position='dodge')

midwest %>% 
    ggplot(aes(state, fill = inmetro, y = poptotal)) + geom_bar(stat='sum') +
    scale_y_continuous(labels=scales::comma)

midwest %>% 
    ggplot(aes(state, fill = inmetro)) + geom_bar(position='fill') +
    scale_y_continuous(labels = scales::percent)

Timeseries

# Build a Time series data set
day <- Sys.Date() - 0:364
value <- runif(365) + seq(-140, 224)^2 / 10000
tsdata <- tibble(day, value)


p <- tsdata %>%
    ggplot(aes(day, value)) + 
    geom_line() +
    scale_x_date(
        #date_labels  = '%Y-%m', 
        date_breaks = '3 months'
        )
p

Interaktivität mit ggplotly

library(plotly)
ggplotly(p)

Interaktivität mit dygraphs

library(dygraphs)

xtsdata <- tsdata %>% 
    as.data.frame() %>% 
    column_to_rownames("day") %>% 
    xts::as.xts()

xtsdata %>% dygraph() %>% dyRangeSelector()